Paper | Slides | Demo

Dodging Dynamical Obstacles Using Turtlebot4 Camera Feed

Overview

This project explores a vision-based approach for Turtlebot4 to dynamically dodge obstacles while navigating toward a designated destination. Unlike previous works that rely on expensive sensors like LiDAR or event cameras, this research focuses on using an affordable stereo camera, the OAK-D Pro, for real-time obstacle detection and avoidance. The implementation includes stereo vision processing, RRT* path planning, Bézier curve smoothing, and a unicycle-based control model.

Features

• Obstacle Detection: Uses OAK-D Pro camera to detect obstacles and determine their coordinates.
• Path Planning: Implements RRT* (Rapidly-exploring Random Trees) to plan collision-free paths.
• Path Smoothing: Uses Bézier curves to create a smooth trajectory for the Turtlebot4.
• Motion Control: Applies a unicycle model for velocity and angular velocity calculations.
• ROS2 Integration: Built using ROS2, allowing modular and scalable implementation.

System Requirements

• Hardware:
  • Turtlebot4
  • OAK-D Pro Camera
  • Raspberry Pi 4B (or compatible computational unit)
• Software:
  • ROS2 (Robot Operating System 2)
  • OpenCV
  • Luxonis DepthAI SDK
  • Python 3

Installation

1. Clone the Repository

$ git clone https://github.com/waynechu1109/Dodging-Dynamical-Obstacles-Using-Turtlebot4-Camera-Feed.git
$ cd Dodging-Dynamical-Obstacles-Using-Turtlebot4-Camera-Feed

2. Set Up ROS2 Environment

Ensure that ROS2 is installed and sourced before running the system:

$ source /opt/ros/humble/setup.bash

Running the Project

1. Start the Camera Node

$ ros2 launch turtlebot4_camera camera.launch.py

2. Run the Obstacle Detection Module

$ ros2 run obstacle_detection detection_node

3. Run the Path Planning and Control Node

$ ros2 run path_planning planner_node

4. Execute the Turtlebot4 Movement

$ ros2 run robot_controller movement_node

Implementation Details

1. Obstacle Detection

• The OAK-D Pro camera provides depth information to estimate obstacle positions.
• ROS2 topics /camera_data and /obstacle_info handle obstacle tracking.

2. Transformation to Odometry Frame

• Converts obstacle coordinates from the camera frame to the robot's odometry frame using transformation matrices.

3. Path Planning with RRT*

• Generates a feasible path avoiding obstacles detected in real-time.

4. Path Smoothing with Bézier Curves

• Smooths the RRT* path to create a natural, drivable trajectory.

5. Motion Control

• Implements a unicycle-based control model to compute velocity and angular velocity.

Results

Simulation

The robot successfully detects moving obstacles, recalculates paths, and dodges them dynamically while progressing toward its destination.

Future Applications

• Self-Driving Cars: Potential to reduce costs by replacing LiDAR-based obstacle avoidance with camera-based systems.
• Unmanned Aerial Vehicles (UAVs): Enhancing drone navigation safety by incorporating vision-based obstacle avoidance.

Contact

Developed by Wei-Teng Chu under the mentorship of Neilabh Banzal, Parth Paritosh, Scott Addams, and Prof. Jorge Cortés at the Multi-Agent Robotics (MURO) Lab, UC San Diego.

For inquiries, feel free to reach out via GitHub Issues or email.